The identification of data objects and their relationships should begin with the definition of a normal entity-relationship model, and the extension of this model to reflect the representations of data objects as well as the logical data entities involved. The objective should be to achieve an optimal balance between the number of object classes and the size and complexity of each class, since each object may itself be composed of other objects, also bearing in mind that multiple objects of the same class may exist. Optimizing the number of object classes will allow the number of message classes to be minimized, which in turn simplifies the design of the methods and the eventual testing of the object classes.